EP0917113B1

EP0917113B1 - Seal detection system and method

Info

Publication number: EP0917113B1
Application number: EP98121376A
Authority: EP
Inventors: Zhigang Fan; John W. Wu; Mike C. Chen
Original assignee: Xerox Corp
Current assignee: Xerox Corp
Priority date: 1997-11-13
Filing date: 1998-11-10
Publication date: 2004-08-25
Anticipated expiration: 2018-11-10
Also published as: EP0917113A3; DE69825842D1; JP2009104663A; EP0917113A2; US6067374A; BR9804607A; DE69825842T2; BR9804607B1; JPH11250260A

Description

Field of the Invention

This invention is generally related to electronic image recognition techniques and, more particularly, to a seal detection system and method that detects and authenticates seals in complex images.

Background of the Invention

The ability to detect seal patterns in an image can be useful in copier machines or scanners for the purpose of authenticating documents or preventing counterfeiting. The challenge of incorporating such a method in current copier or scanning technology is the difficulty with detecting seals patterns in a rotation or shift invariant manner. Specifically, the pattern could be of any orientation and at any location of the image. The orientation and the location of the seal can be relatively simple to estimate in the case of a single seal within a plain background; however, it becomes a major obstacle when the seals are embedded in some complicated image background.
The US Patent No 5 533 144 discloses an anti-counterfeit detector and method which identifies whether a platen image portion to be photocopied contains one or several note patterns. An off-line training is performed by sampling suspect currencies and recording said samples to use them as templates, the detection is then performed in a rotation and shift invariant manner. Specifically, the pattern can be of any orientation and at any location of the image and can be embedded in any complicated image background. The image to be tested is processed block by block. Each examined block goes through a smoothing to see if it possibly contains a pixel intensity orientation that corresponds to a preselected anchor point on the template. The orientation of the edges (anchor points) potentially contained in the blocks is then estimated. Finally, for a potential anchor point, a matching procedure is performed against stored templates to decide whether the pre-selected monetary note patterns are valid once detected.

Summary of the Invention

It is an object of the present invention to improve counterfeit detection to detect distinctive marks in documents. This object is achieved by providing a counterfeit detection method according to claim 1 and a counterfeit detection system according to claim 3.
A detection system and method that detects distinctive marks, such as seals or other patterns, in images for purposes of authentication or to defeat counterfeiting is presented. This detection method has the ability to identify whether an image contains one or several pre-selected distinctive marks.
A detector is first trained off-line with smoothed examples of the distinctive marks of interest to be detected during operation. The distinctive marks are each stored as templates. After training, to detect marks, a four step procedure consisting of binarization, location estimation, orientation estimation and template matching is performed. Binarization extracts a binary bitmap from the input image. A pixel in the bitmap is set to be "1" if the color of the corresponding pixel in the input image is close to the color of the template to be matched to the input image. Location estimation detects the "suspects", or the potential mark patterns, and estimates their location. The relative orientation of the suspects and the template is then evaluated, so they can be aligned (this method is rotation and shift invariant). Finally, after orientation, the suspect and template are compared and analyzed to verify if suspect is legitimate. A suspect mark can be in any orientation and at any location within an image.
According to a first aspect of the invention, the method can be summarized as follows:
a detector is trained off-line with smoothed distinctive marks resulting in templates which are generated and recorded for each of the distinctive marks;
sample images bearing suspect marks are received by the detector and the location and orientation of the suspect marks are identified;
the templates are rotated and shifted for alignment of he templates to the suspect marks;
the templates and the suspects marks are compared to determine whether there is a match.

According to third aspect of the invention, the method can be carried out in a system comprising a microprocessor programmed to become familiarized with a plurality of seals through training and to analyze and detect distinctive marks within tested documents. A memory is used to store a smoothed version of the marks of interest. A scanner may be used during training and detection to accept training marks and images bearing suspect marks, and transmits the captured images to the microprocessor; however, digitized representations of the training marks and images may also be accepted electronically over networks.
In a preferred embodiment is a microprocessor-based document processing system wherein a microprocessor is programmed to
detect control marks found on controlled documents, and
suspend further action of suspect documents not bearing said control marks
In a further preferred embodiment the system comprises an indicator means for indicating whether said control marks and said suspect marks of said suspect document match.
In a further preferred embodiment the output from said indicator means is used by said system to facilitate further action on said suspect document.
Other advantages and salient features of the invention will become apparent from the detailed description which, taken in conjunction with the drawings, disclose the preferred embodiments of the invention.

Description of the Drawings

Figure 1 is an illustration of a matched filter applied by the system to detect the presence of any suspects;
Figure 2 illustrates the detection starting from the left boundary of the original bitmap for a mark at the fine resolution (a search is conducted from left to right in two nxn blocks, which are m blocks away from the location of the strong peak);
Figure 3 illustrates a gray map on a circle of radius c with which data are sampled;
Figure 4 illustrates a peak for the sample mark as "A";
Figure 5 illustrates a peak for the template as "B"; and
Figure 6 is an block diagram of the system used to carry out the training and detection method of the invention.

Detailed Description of the Invention

"Seal" will be used throughout the balance of this disclosure to define distinctive marks and distinctive patterns which may be commonly used in the document authentication art.
The detector is first trained off-line with examples of the seals to be detected. Training is conducted by scanning seals into a microprocessor-based detection system using scanning techniques known in the art. The seals are converted into smoothed templates representing each respective seal. The training specific to this invention occurs after the system has received the electronic representation of the seals and consists of two steps. First, the color of the seal template is recorded. Second, the seal template is smoothed using an averaging filter (the same filter used in detection). The results, a smoothed version of the binary of the seal patterns, are recorded as a template.
To detect each seal, a four step procedure consisting of binarization, location estimation, orientation estimation and template matching is performed. Binarization extracts a binary bitmap from the input image. A pixel in the bitmap is set to be "1" if the color of the corresponding pixel in the input image is close to the color of the seal to be detected. Location estimation detects the "suspect", or the potential seals, and estimates their location. The relative orientation of the suspect and the seal is then evaluated, so they can be aligned. Finally, a template match verifies if the candidate is really the seal to be detected.
The location estimation is performed in two resolution. The detection of the suspects and the estimation of their rough positions are followed by a refinement of the locations. First, a low resolution version of the bitmap is produced. Each nxn pixels in the original bitmap is reduced to one pixel, which is set to be "1" if at least on of the nxn pixels is "1". A matched filter is then applied to detect the presence of any suspects. The kernel of the filter is given in Figure 1. The strong peaks in the filtering result indicate the rough locations of the centers of the suspects. Once a strong peak is detected, the left, right top and bottom boundaries are searched in the original bitmap. Figure 2 illustrates the detection of the left boundary at the fine resolution. A search is conducted from left to right in two nxn blocks, which are m blocks away from the location of the strong peak, where m = r/n and r is the radius of the seal to be detected. The first column which contains at least one "1" pixel gives the left boundary. The right, top and bottom boundaries can be obtained in a similar fashion. The x and y-coordinates of the center of the suspect are estimated as, x0 = (left boundary + bottom boundary)/2 and y0 = (top boundary + bottom boundary)/2, respectively.
The data in the window, centered at (x0,y0) as shown in Figure 1, are smoothed using an averaging filter to create a gray map. The actual window size is slightly larger than the diameter of the tested mark. A high (low) pixel value in the gray map corresponds dense "1" ("0") pixels in the bitmap. For the areas where "1" pixels and "0" pixels intermingle, a gray value in the middle results. This gray map is used for orientation estimation and template matching by comparing it to the gray map obtained from the mark to be detected.
Referring to Figures 3, data are sampled in the gray map on a circle of radius c. The highest peak (or the lowest valley) position of the data reveals the orientation. Features other than the peak or valley position, or a transformation of the original data can also be used to determine the orientation. Figure 4 illustrates a peak for the sample mark as "A". Figure 5 illustrates a peak for the template as "B". A difference in rotation is noticeable upon comparing the peaks of the two sequences of data, sample (Figure 4) and template (Figure 5). To accomplish alignment, the template must be rotated "RR", as shown in Figure 3, so that the peak of the template "B" matches the peak "A" of the sample.
Once the orientation of a suspect is determined, the template, which is the smoothed version of the seal bit pattern is rotated to align with the suspect. A template matching can be performed as revealed in US Patent No. 5,533,144 to Fan, or by using any other standard techniques.
Referring to Figure 5, the detection method can be carried out in a system 11 comprising a microprocessor 14 programmed to become familiarized with a plurality of seals through training and to analyze and detect seals within tested documents. A memory 13 is used to store the seals of interest works hand in hand with the microprocessor 14 during detection. A scanner 12 is used with the system during training and detection to accept seals and images bearing seals (referred to as a "Test Image" in the figure) and transmit the seals and images to the microprocessor; however, the seals and images may also be transmitted electronically over networks, rather than directly from a scanner. After processing through the microprocessor 14, a testing result is "Output' to indicate counterfeit testing results. The output can be used by controlled systems, such as copiers and scanners, to suspend further action on documents where counterfeiting is suspected. It is noted that the microprocessor may be replaced by hardware equivalents through technical methods know in the art.

Claims

A counterfeit detection method that detects distinctive marks in documents, comprising steps of:

training a detector off-line with distinctive marks resulting in templates which are generated and recorded for each of said distinctive marks;

receiving sample images bearing suspect marks by said detector and identifying the location and orientation of said suspect marks on said sample images;

rotating and shifting said templates for alignment of said templates to said suspect marks; and

comparing said templates and said suspects marks to determine whether there is a match

characterized in that
said step of training further comprises recording a color of said distinctive marks during training and smoothing said distinctive marks using a binary averaging means, whereby said color of said distinctive marks and said smoothed version of the binary averaging means of said distinctive marks are generated and recorded as said templates.
The method of claim 1, wherein a result is generated by comparing said templates and said suspects marks to determine whether there is a match, and said result is utilized to facilitate further action on said sample images.
A counterfeit detection system, comprising:

a scanning means for capturing distinctive marks during training and related marks during detection;

a mircoprocessor;

means for transmitting said marks to the microprocessor;

a memory for recording said distinctive mark;

whereby the microprocessor comprises:

means for generating templates for each of said distinctive marks,

means for analyzing and detecting related marks within tested documents,

characterized in that
said templates comprise a color of said distinctive marks provided by using a binary averaging means.
The system of claim 3, further comprising a signal means for indicating results of said analysis.
The system of claim 4, wherein an output by said signal means is used by electronic document handling system to facilitate further action on said tested documents.
The system of claim 5, wherein said signal means output can be used by controlled systems, such as copiers and scanners, to suspend further action on documents where counterfeiting is suspected.